ProLiant DL740 tames enterprise apps

When an ordinary Intel-based server doesn’t deliver the oomph or reliability that a specific application needs, you have three choices. One, you can scale vertically by migrating to a server that has a similar architecture but provides more capacity and high-availability features. Two, you can scale horizontally by establishing a cluster of small servers, the ultimate example being a blade-server system. Or three, you can move the application to a different architecture such as a high-end Sun server or an IBM mainframe.

Of all those options, vertical scalability is generally the simplest because applications typically must be reconfigured, rewritten, or even redesigned to run in a clustered environment or on a different class of computer, such as that mainframe. That’s not only complex, it’s costly and time-consuming. Also, adding more servers increases ongoing management and administrative costs.

If you’re making the effort to move an application to a bigger server, it must be big enough and reliable enough to make the migration worthwhile. That’s what Hewlett-Packard has delivered with its ProLiant DL740, an eight-processor, Xeon MP-based server with just about every high-end reliability feature imaginable in a high-powered x86-based server. It won’t give you the high-end reliability of a mainframe or Tandem box, but the DL740 is an impressive platform for vertical scalability.

Spectacular specs

The ProLiant DL740 is a powerhouse of a server. The spec list alone is formidable with four 1.5GHz or 2GHz Xeon MP processors (expandable to eight), 4GB of memory (expandable to 64GB), eight hot-swap PCI-X expansion slots, two integrated Gigabit Ethernet NICs, an onboard management processor, an integrated RAID controller, and four hot-swap Ultra320 SCSI hard drives (see “Xeon vs. Xeon” in the related info box that accompanies this article at right).

For high availability beyond storage RAID, the server includes two redundant hot-swap power supplies (removable from the front of the server) and two hot-swap cooling fans, as well as the so-called RAID memory. That technology uses the same concept as RAID storage, striping RAM across five hot-swap cartridges. Any one cartridge can be removed and replaced while the server is running, without disturbing the operating system or applications. In fact, HP claims that memory can even be increased dynamically by swapping in larger cartridges, although it’s very likely that the server’s operating system and applications may not recognize that new memory until you reboot.

Indeed, when we got past the trauma of yanking a memory cartridge out of a running server, we were impressed. It’s a clever concept and is well-implemented except for the fact that the RAID memory cartridges can be reached only by removing the server’s cover. On our “bread-rack” test system, that presented no inconvenience, but in a normal deployed situation, with the server loaded into a standard rack, memory swapping would be a challenge. In the ProLiant DL760G2, the DL740’s big brother, the cartridges are removable from the front of the server.

We found ProLiant DL740’s performance, scalability, and ease of management to be excellent. The review system HP provided was a robust configuration, with eight 2GHz processors, each with 2MB of L3 cache (strongly recommended over the 1.5GHz Xeon MP processors with only 1MB cache); 4GB RAM in five cartridges of 1GB each; and four 36GB Ultra320 drives. All worked perfectly with our tests under Windows 2000 Datacenter Server, including the RAID memory.

Radical RAID

For our review system, HP upgraded the server’s RAID controller and activated the server’s onboard iLO (Integrated Lights-Out) management processor — a $399 option. Frankly, it’s something that should be included at no extra charge on a server with a base price of $45,000 (ours was buffed up to $78,000). There’s nothing new about iLO, by the way; it’s the same management chip that Compaq has been shipping for years and it does a good job providing remote diagnostics, console, and other functions, even if the server itself is down.

If it seems like we’re making the RAID memory out to be a big deal, well, we are. Memory chips, similar to processors, Ethernet NICs, and other server components that lack moving parts, are very reliable today. The weakest spots in a server are the power supplies, fans, and rotating media. So, for most servers, RAID memory wouldn’t be necessary. But in a true enterprise-class server — one that might be running many mission-critical applications — the goal is to attain as close to 100 percent uptime as possible.

Striping memory across multiple modules ensures that if one fails, the system doesn’t die; that’s the largest benefit of the RAID memory. The hot-swap capability is icing on the cake. Things being equal, a system with RAID memory would be preferred over one with normal ECC (error-correcting code) memory. ECC RAM can detect and correct single-bit errors during memory operations, as well as some other transient memory faults, but it can’t recover from a memory-chip failure.

(Note that IBM has recently introduced the capability to interconnect two of its eight-way eServer xSeries 440 servers together and calls the result a 16-processor server. We haven’t examined or tested this configuration, which also runs 2GHz Xeon MP chips. According to IBM’s spec sheets, the xSeries 440 uses ECC RAM.)

The DL740 is an outstanding server, offering top-shelf processing, infrastructure, and high-availability functions for vertically scaling enterprise applications — think WebSphere, DB2, WebLogic, Oracle, Siebel, and SAP. Plus for those sorts of applications, it’s much simpler and easier to administer one (or a small number) of high-end servers than a bunch of small ones. But for mission-critical, processor-intensive apps, or for consolidation of many smaller servers into one, this system is the one to beat.